MATH
Harbor adapterMATH: 竞赛级数学（AMC/AIME等）
- Samples
- 5,000
- Released
- 2021
- Scores
- 46
How to run
ltb · math-500
$ npm i -g lettokenburn-cli$ ltb bench pull math-500 --n 50$ ltb bench run ./math-500 -a claude-code -m anthropic/claude-opus-4-8$ ltb bench push
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
9 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Kimi K2.5 | 98.0 | N/A | Self-reported | low | Jun 1, 2026 | kaggle.com |
| Claude Sonnet 4.6 | 97.8 | N/A | Self-reported | low | Jun 1, 2026 | kaggle.com |
| Claude Opus 4.6 | 97.6 | N/A | Self-reported | low | Jun 1, 2026 | kaggle.com |
| DeepSeek R1 | 97.3 | N/A | Self-reported | low | Jun 1, 2026 | kaggle.com |
| MiMo-V2-Flash | 94.1 | N/A | Self-reported | low | Jun 1, 2026 | kaggle.com |
| Gemini 3.1 Pro | 94.0 | N/A | Self-reported | low | Jun 1, 2026 | kaggle.com |
| Mistral Large | 93.6 | N/A | Self-reported | low | Jun 1, 2026 | kaggle.com |
| Gemini 3 Pro | 91.0 | N/A | Self-reported | low | Jun 1, 2026 | kaggle.com |
| GLM-5 | 88.0 | N/A | Self-reported | low | Jun 1, 2026 | kaggle.com |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsLoading community runs…
About
MATH-500 (MATH 500子集) — 500道精选竞赛数学题. Scoring: 准确率. Difficulty: 高（趋于饱和）. Imported from benchmark_research.md (2026-06).